etcdserver: adjust election ticks on restart

Signed-off-by: Gyuho Lee <gyuhox@gmail.com>

etcdserver/raft.go

@@ -97,6 +97,7 @@ type raftNode struct {
 	term         uint64
 	lead         uint64
 
+	tickMu sync.Mutex
 	raftNodeConfig
 
 	// a chan to send/receive snapshot
@@ -151,6 +152,13 @@ func newRaftNode(cfg raftNodeConfig) *raftNode {
 	return r
 }
 
+// raft.Node does not have locks in Raft package
+func (r *raftNode) tick() {
+	r.tickMu.Lock()
+	r.Tick()
+	r.tickMu.Unlock()
+}
+
 // start prepares and starts raftNode in a new goroutine. It is no longer safe
 // to modify the fields after it has been started.
 func (r *raftNode) start(rh *raftReadyHandler) {
@@ -163,7 +171,7 @@ func (r *raftNode) start(rh *raftReadyHandler) {
 		for {
 			select {
 			case <-r.ticker.C:
-				r.Tick()
+				r.tick()
 			case rd := <-r.Ready():
 				if rd.SoftState != nil {
 					newLeader := rd.SoftState.Lead != raft.None && atomic.LoadUint64(&r.lead) != rd.SoftState.Lead
@@ -376,7 +384,7 @@ func (r *raftNode) resumeSending() {
 // speeding up election process.
 func (r *raftNode) advanceTicks(ticks int) {
 	for i := 0; i < ticks; i++ {
-		r.Tick()
+		r.tick()
 	}
 }
 

etcdserver/server.go

@@ -530,12 +530,51 @@ func NewServer(cfg ServerConfig) (srv *EtcdServer, err error) {
 	return srv, nil
 }
 
+func (s *EtcdServer) adjustTicks() {
+	clusterN := len(s.cluster.Members())
+
+	// single-node fresh start, or single-node recovers from snapshot
+	if clusterN == 1 {
+		ticks := s.Cfg.ElectionTicks - 1
+		plog.Infof("%s as single-node; fast-forwarding %d ticks (election ticks %d)", s.ID(), ticks, s.Cfg.ElectionTicks)
+		s.r.advanceTicks(ticks)
+		return
+	}
+
+	// retry up to "rafthttp.ConnReadTimeout", which is 5-sec
+	// until peer connection reports; otherwise:
+	// 1. all connections failed, or
+	// 2. no active peers, or
+	// 3. restarted single-node with no snapshot
+	// then, do nothing, because advancing ticks would have no effect
+	waitTime := rafthttp.ConnReadTimeout
+	itv := 50 * time.Millisecond
+	for i := int64(0); i < int64(waitTime/itv); i++ {
+		select {
+		case <-time.After(itv):
+		case <-s.stopping:
+			return
+		}
+
+		peerN := s.r.transport.ActivePeers()
+		if peerN > 1 {
+			// multi-node received peer connection reports
+			// adjust ticks, in case slow leader message receive
+			ticks := s.Cfg.ElectionTicks - 2
+			plog.Infof("%s initialzed peer connection; fast-forwarding %d ticks (election ticks %d) with %d active peer(s)", s.ID(), ticks, s.Cfg.ElectionTicks, peerN)
+			s.r.advanceTicks(ticks)
+			return
+		}
+	}
+}
+
 // Start performs any initialization of the Server necessary for it to
 // begin serving requests. It must be called before Do or Process.
 // Start must be non-blocking; any long-running server functionality
 // should be implemented in goroutines.
 func (s *EtcdServer) Start() {
 	s.start()
+	s.goAttach(func() { s.adjustTicks() })
 	s.goAttach(func() { s.publish(s.Cfg.ReqTimeout()) })
 	s.goAttach(s.purgeFile)
 	s.goAttach(func() { monitorFileDescriptor(s.stopping) })